Automatic detection and classification of landslides using AI model and post-event optical satellite imagery

2023 Fiscal Year Final Research Report

• Project/Area Number: 21K04606
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Review Section: Basic Section 25030:Disaster prevention engineering-related
• Research Institution: Chiba University
• Principal Investigator: Liu Wen 千葉大学, 大学院工学研究院, 准教授 (60733128)
• Project Period (FY): 2021-04-01 – 2024-03-31
• Keywords: 土砂災害 / 機械学習 / 衛星光学画像

Outline of Final Research Achievements

We aligned the landslide data due to the July 2018 torrential rainfall in Hiroshima Prefecture, and that due the 2018 Hokkaido Eastern Iburi earthquake in Hokkaido Prefecture. Using machine learning techniques, two predictive models for landslides were developed. These models incorporated various types of data such as elevation models, geological information, and either rainfall or seismic motion data. When tested on the respective disasters, these models demonstrated an overall accuracy of approximately 75%.
Additionally, a deep learning model was created to identify the shape and extent of landslide areas using post-event satellite optical images. This model successfully detected 77% of the landslide areas with an accuracy rate of 92%.

Free Research Field

防災工学

Academic Significance and Societal Importance of the Research Achievements

本研究では，2018年7月の広島県豪雨と2018年北海道胆振東部地震による土砂崩壊箇所のデータを整備し，それぞれの災害を予測するモデルを構築した．これらの研究成果は査読雑誌に掲載された．作成した土砂災害輪郭データを広く公開することで，今後の土砂災害発生要因の解明と機械学習などの検出手法の発展に貢献する．
また，土砂崩壊域の形状・範囲を検出できる深層学習モデルも構築した．地域安全学会と土木学会年次に成果を発表する．このモデルは土砂移動の要因を考慮せず，事後光学衛星画像のみを使用して土砂崩壊を検出するもので，大規模災害時の情報収集・緊急対応に大きく貢献するものと期待できる．